Input: Graph dataset G = {G₁, G₂, ..., Gₙ}
Output: Graph-level classification predictions Y_pred
1: Initialize model parameters θ
2: // Main training loop
for each epoch do
3: // Iterate over each graph in the dataset
for each graph Gᵢ in G do
4: // Step 1: Hypergraph Construction
E_intra, E_inter = generate_hyperedges(Gᵢ)
5: // Step 2: Hypergraph Convolutions
H_node, H_edge = hypergraph_convolution(E_intra, E_inter)
6: // Step 3: Context-Aware Attentive Graph-Level Pooling
H_graph = attentive_pooling(H_node, H_edge)
end for
end for
7: // Step 4: High-Order-Aware Graph Augmentation
G_augmented_list = high_order_augmentation(G)
8: // Step 5: Self-Supervised Contrastive Learning
for each G_aug in G_augmented_list do
9: H_aug_node, H_aug_edge = hypergraph_convolution(E_intra, E_inter)
10: H_aug_graph = attentive_pooling(H_aug_node, H_aug_edge)
11: Loss = contrastive_loss(H_graph, H_aug_graph)
12: θ = optimizer.update(Loss, θ) end for
13: // Step 6: Graph-Level Classification
Y_pred = []
for each test graph G_test in G_test do
14: E_intra_test, E_inter_test = generate_hyperedges(G_test)
15: H_node_test, H_edge_test = hypergraph_convolution(E_intra_test, E_inter_test)
16: H_graph_test = attentive_pooling(H_node_test, H_edge_test)
17: Y_pred_test = classifier(H_graph_test)
18: Y_pred.append(Y_pred_test) end for
19: return Y_pred